The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various framework and utility CLI tools to scaffold projects, including bun create next-app, bun create astro, composer create-project roots/bedrock, and bun create expo-app.
[EXTERNAL_DOWNLOADS]: The skill fetches project templates and official scaffolds from multiple external sources, including GitHub repositories and NPM registries. Notable downloads include the Aria Builder official repository (github.com/ariabuilder/aria.git), the Atomic Payload template (@pro-laico/create-atomic-payload), and various CMS/e-commerce CLIs like create-wolly, create-emdash, create-tina-app, create-medusa-app, and create-puck-app.
[INDIRECT_PROMPT_INJECTION]: The skill implements a 6-stage interactive interview process that ingests user-provided purpose, audience, and scope data. This data is subsequently interpolated into project configuration files, governance documents (AGENTS.md), and context files (.agents/context/*.md). While the skill instructs the agent to use 'real content only', this ingestion surface presents a minor risk for indirect instruction injection via user responses.
Ingestion points: Interactive onboarding interview (Stage 1 & 2) prompts described in SKILL.md.
Boundary markers: Stage 5 instructions command the agent to grounded writing from 'real employee answers' and to use an 'Exclusion List' section to limit scope.
Capability inventory: The skill uses bash for command execution and write_to_file for creating project files.
Sanitization: No explicit sanitization or escaping of user-provided strings is mentioned in the prompt generation logic.
[DYNAMIC_EXECUTION]: The skill dynamically generates several executable assets and configuration files at runtime, including a Git pre-commit.sh hook for secret defense, relational database schemas (src/lib/schema.ts), and various platform-specific configurations (wrangler.toml, vercel.json, capacitor.config.ts).

new-project